Discussion

Approximately 80% of patients diagnosed with head and neck cancer (HNC) receive radiotherapy (RT) at some point during the disease [66]. RT remains a cornerstone non-surgical treatment modality, balancing oncologic control with treatment-related toxicity [67].

Conventional radiotherapy delivers multiple beams from different angles based on two-dimensional fluoroscopic simulation. Although effective, this approach is associated with substantial toxicity due to unavoidable irradiation of adjacent healthy tissues [68]. Cobalt-60 teletherapy, one of the earliest forms of external beam RT, utilizes high-energy gamma radiation emitted by the radioactive isotope Cobalt-60. Advances in radiation planning and delivery have led to the incorporation of three-dimensional conformal radiotherapy (3D-CRT), intensity-modulated radiotherapy (IMRT), and intensity-modulated proton therapy (IMPT), enabling improved dose conformity and reduced exposure of surrounding structures [67].

In HNC, total doses typically range from 50 to 70 Gy, administered in conventional or altered fractionation schedules [69]. ORNJ appears to be more prevalent in patients treated with IMRT than in those who received 3D-CRT [70]. Furthermore, IMPT has been reported to reduce the likelihood of ORNJ development by 2% compared with conventional IMRT [69]. However, a recent study comparing the incidence of ORNJ in patients with oropharyngeal squamous cell carcinoma treated with IMPT or IMRT found a higher rate of ORNJ in the IMPT group [71]. Overall, these findings indicate that, despite the potential dosimetric advantages of proton therapy, the association between radiotherapy modality and ORNJ risk remains controversial and should be interpreted with caution.

Brachytherapy allows localized delivery of high radiation doses with steep dose gradients, theoretically minimizing collateral tissue damage [72]. It is often indicated for early-stage tumors (T1/T2). However, Danielsson [72] reported an increased risk of osteoradionecrosis of the jaws (ORNJ) associated with brachytherapy without a corresponding survival benefit [72]. Consistently, our meta-analysis demonstrated increased odds of ORNJ in patients with T2 (OR = 1.78 [1.32–2.41]) and T4 tumors (OR = 1.98 [1.38–2.84]), suggesting that tumor extent and radiation burden may critically influence complication risk.

The dose–volume relationship remains insufficiently characterized but appears central to ORNJ pathogenesis despite the irradiation modality employed for cancer treatment. Higher radiation doses and larger irradiated bone volumes are consistently associated with increased risk [54]. Reported prevalence ranges from 0% to 20%, reflecting substantial heterogeneity across studies [62].

Dental extraction emerged as the strongest risk factor identified in this systematic review (OR = 12.31 [6.40–23.69]). The risk was elevated both before (OR = 1.15 [1.01–1.31]) and particularly after RT (OR = 8.38 [7.20–9.77]. Moharrami et al. [73] developed and validated an individualized risk prediction model for ORN in patients with head and neck cancer, incorporating clinical, dosimetric, and sociodemographic predictors. Although pre-radiotherapy extractions were not retained as a key predictor and did not improve model performance when experimentally included, the final model identified primary tumor site, mandibular D10cc, smoking pack-years, periodontal condition, and dental insurance status as relevant predictors of ORN risk. These findings suggest that clinical decision-making should not rely solely on average population-level estimates, but rather on individualized risk assessment that considers the interaction between local, systemic, and treatment-related factors [73].

Radiation-induced hypovascularity, hypocellularity, and fibrosis impair alveolar healing, predisposing to chronic non-healing wounds. Additionally, obliteration of marrow vasculature, including branches of the inferior alveolar artery, compromises bone vitality [58]. When osteotomy is performed and the alveoli receives high doses of radiation, the risk of developing ORNJ increases further [74]. Although atraumatic extraction techniques with primary closure have been proposed to mitigate risk, robust evidence supporting their effectiveness remains limited [75]. Jawbone homeostasis is maintained through continuous bone remodeling, in which osteoclast-mediated resorption is tightly coupled with osteoblast-mediated bone formation. Osteocytes act as mechanosensors and regulate this process in response to functional loading, particularly masticatory forces, thereby preserving bone structure, mineral homeostasis, and mechanical competence [76]. Accordingly, any condition that disturbs the balance of bone remodeling might increase the risk of ORNJ.

Dental caries and periodontitis were also identified as relevant contributors. Hyposalivation following RT alters the oral microbiome and reduces salivary protective function, increasing cariogenic potential [77]. Radiation may further compromise dentin microhardness and disrupt the dentin–enamel junction, facilitating bacterial colonization. Periodontitis, frequently present prior to RT, may be exacerbated during treatment and maintain chronic inflammatory foci that increase ORNJ susceptibility [47, 78, 79]. Notably, good oral hygiene demonstrated a protective effect (OR = 0.40 [0.21–0.76]), underscoring the importance of preventive protocols.

Tumor location significantly influenced risk. Mandibular involvement, particularly in cases of bone infiltration, exposes highly irradiated cortical structures with limited vascular reserve. Our findings showed increased odds for oral cavity tumors (OR = 4.00 [2.90–5.51]) and oropharyngeal cancer (OR = 2.34 [1.63–3.34]), reinforcing the role of anatomical and vascular vulnerability. Hyperbaric oxygen therapy (HBOT) remains controversial. Although it increases tissue oxygenation and may promote angiogenesis, reported resolution rates are modest (~ 15%), and existing trials suffer from methodological limitations [9]. Its routine use therefore remains unsupported by high-level evidence.

Chemotherapy, particularly cisplatin-based regimens, may potentiate radiation-induced tissue damage through radiosensitization (100 mg/m² ≈ 7.2 Gy equivalent) [80]. This finding is consistent with the results of our meta-analysis, which suggested that chemotherapy may be associated with a reduced risk of ORNJ development. Further studies are needed to clarify the biological mechanisms underlying this association. Nevertheless, previous studies have investigated biomarkers related to chemotherapy response in an attempt to predict ORNJ development. It has been observed that patients who exceed 285,000 cells/µL of platelet had an increased risk of developing ORNJ in patients with nasopharyngeal carcinoma treated with chemoradiotherapy, whereas patients with platelet counts equal to or below 285,000 cells/µL had a lower risk of developing the disease. These results were evaluated only at baseline platelet levels, without considering their variation throughout treatment, which may limit the definition of an optimal cutoff point for ORNJ [81].

Other relevant factors, including blood flow, tissue oxygenation, and cytokine expression, were not assessed [81]. Furthermore, no sample size or power calculation was performed. Hematological parameters, including anemia, have also been described as relevant predictors of ORNJ [81]. Among these, the lymphocyte-to-monocyte ratio (LMR) has been proposed as a potential biomarker for predicting ORNJ in patients with OSCC undergoing radiotherapy. Accordingly, LMR-guided pre-treatment dental extractions may contribute to ORNJ prevention [82]. In addition, sclerostin, a glycoprotein encoded by the SOST gene and primarily expressed by osteocytes, has been implicated in bone metabolism. As a potent inhibitor of bone formation, sclerostin antagonizes the canonical Wnt/β-catenin pathway, thereby modulating osteoblast activity. Consequently, increased sclerostin expression may impair osteoblastic differentiation and promote apoptosis [83].

In individuals with poorly controlled type 1 diabetes mellitus, impaired bone formation may lead to delayed healing and deficient tissue repair [84]. The early phase of bone healing in patients with diabetes is characterized by reduced osteoid matrix production and decreased cellularity, which may be attributed to impaired recruitment of mesenchymal stem cells (MSCs), together with reduced MSC proliferation and osteogenic differentiation. In addition, diabetes may promote aberrant activation of the NF-κB pathway, thereby exacerbating inflammation, at least in part through a reduction in immunomodulatory MSCs [84].

Persistent inflammation, marked by increased TNF levels, may impair bone healing by reducing MSC expansion through FOXO1 upregulation and IHH suppression [84]. The relationship between hyperglycemia and bone metabolism remains complex and not fully elucidated. Although insulin has anabolic effects on bone tissue, diabetes mellitus may paradoxically impair bone remodeling by decreasing osteoblast proliferation and survival, while enhancing osteoclastic activity through increased RANK expression. In addition, the accumulation of advanced glycation end products (AGEs) may compromise collagen quality and bone strength, thereby further delaying tissue repair [84]. Despite strong biological plausibility, diabetes did not demonstrate statistically significant association with ORNJ in our meta-analysis analysis.

Hypertension has also been suggested to impair alveolar bone healing [84]. Osteoprotegerin (OPG), receptor activator of nuclear factor kappa-B (RANK), and receptor activator of nuclear factor kappa-B ligand (RANKL) are central regulators of bone remodeling. OPG functions as a decoy receptor for RANKL, preventing its binding to RANK on pre-osteoclasts and thereby inhibiting osteoclast differentiation and bone resorption. In contrast, RANKL promotes osteoclastogenesis and osteoclast-mediated resorptive activity. Accordingly, the OPG/RANKL ratio reflects the balance between bone formation and bone resorption, which is critical for tissue repair. Disruption of this axis in hypertensive individuals may contribute to delayed alveolar bone healing [84].

Smoking and alcohol consumption further compromise wound healing. Nicotine can affect cellular protein synthesis, reducing the adhesive capacity of gingival fibroblasts and impairing tissue repair [85]. In addition, tobacco exposure promotes pro-inflammatory cytokine expression, endothelial dysfunction, and reduced fibroblast adhesion, thereby compromising vascular supply and wound healing [85, 86]. Alcohol consumption inhibits osteoblast proliferation and angiogenesis, while also disrupting early inflammatory signaling and collagen synthesis [87, 88].

Recent literature has introduced more contemporary approaches to ORNJ classification. Watson et al. [63] proposed a clinical-radiographic classification system, termed ClinRad, designed to stratify patients according to disease severity and risk of progression. In this study, a high-risk group was identified, with 5.7% of patients progressing to the most severe form of ORNJ. The ClinRad system classifies ORNJ into progressive stages: stage 0, defined by radiographic changes without clinical bone exposure; stage 1, characterized by bone exposure limited to the alveolar bone; stage 2, involving the basal bone or maxillary sinus; and stage 3, representing advanced disease with pathological fracture, fistula, or oroantral/oronasal communication. Patients diagnosed at stage 3 were more likely to develop fractures or require earlier surgical resection compared with those in lower stages. Despite its superior performance compared with other classification systems, the model remains limited by its retrospective single-center design, the absence of detailed data on the actual IMRT dose delivered, and the lack of external validation to date [63].

A recent interobserver diagnostic performance study assessed the accuracy and reproducibility of ORNJ detection and staging using CT and panoramic radiography [89]. Although the combination of CT and panoramic radiography improved diagnostic performance, interobserver agreement for ClinRad remained limited. These results indicate that, despite advances in imaging-based classification, substantial variability persists among observers, which may compromise consistent staging and clinical decision-making in ORNJ.

Humbert-Vidan et al. [90] investigated clinical and dosimetric factors using machine learning approaches to develop predictive models for ORNJ. The study included 96 patients, comprising 48 individuals with ORNJ and 48 controls. The variables analyzed encompassed clinical factors, including dental extractions, surgical interventions, and lifestyle habits, as well as radiotherapy-related dosimetric parameters [90]. Several machine learning algorithms were tested, including multivariate logistic regression (LR), support vector machine (SVM), random forest (RF), adaptive boosting (AdaBoost), and artificial neural networks (ANN). Among the models evaluated, ANN achieved the highest accuracy (77%), followed by SVM (76%), AdaBoost (75%), and LR (75%), although no statistically significant differences were observed among them [90]. Overall, these findings indicate that machine learning-based models may predict ORNJ occurrence with moderate accuracy and support a more comprehensive risk assessment by integrating multiple clinical and dosimetric variables [90].

Another study applying machine learning methods showed that a parsimonious Random Survival Forest (RSF) model, based on relevant clinical and dosimetric predictors, can reliably estimate the individualized risk of ORNJ in patients with head and neck cancer [73]. By accounting for the competing risk of death, the model minimizes risk overestimation, a limitation commonly associated with traditional predictive approaches. Its robust performance and interpretability support its potential clinical applicability [91]. In addition, the development of an interactive web-based tool may facilitate its integration into clinical practice, supporting personalized treatment planning and decision-making after radiotherapy, with the goal of reducing ORNJ risk [73].

Chen et al. [91] used causal machine learning to demonstrate that dose-response parameters exert a causal influence on ORNJ development, supporting the presence of a dose-response relationship. The authors also identified heterogeneity in these effects, with a stronger impact among patients aged 50–60 years and a weaker effect among those older than 70 years. These findings suggest that individualized radiotherapy planning, particularly when incorporating age-based risk stratification, may help reduce ORNJ risk and underscore the potential of causal inference approaches to support more precise clinical recommendations [91].

The mandible is the most frequently affected anatomical site by ORNJ, likely because of its relatively limited vascular supply, thin mucosal coverage, mechanical stress from mastication, and high remodeling demand [88]. Our systematic review findings indicate that early radiographic changes are particularly evident in the posterior mandible, characterized by atypical bone resorption and sclerosis along non-cortical margins. Radiation-induced endothelial injury promotes progressive microvascular obliteration, chronic hypoxia, and impaired bone remodeling, all of which represent central mechanisms in ORNJ pathogenesis [88].

Magnetic resonance imaging demonstrated 92% specificity for identifying soft tissue alterations and differentiating malignant features. Compared with CBCT, MRI offers advantages in the assessment of vascular characteristics and structural vascular damage, whereas CBCT remains superior for evaluating bone structures [92]. Black bone MRI has emerged as a promising high-resolution alternative to CT for measuring mandibular cortical bone, with potential applicability in the monitoring and early diagnosis of ORNJ. Despite its inherent limitations, MRI is particularly relevant because it is a non-ionizing imaging modality, allowing serial and more frequent assessment of changes in cortical bone thickness [93].

Scintigraphy has been proposed as a useful imaging modality for the early diagnosis of ORNJ, with a reported sensitivity of 100%. By reflecting changes in bone metabolism and blood flow, it allows the detection of increased osteoblastic activity in affected mandibulares sites [72, 94]. PET/CT, in turn, is valuable for identifying inflammatory soft tissue changes and tumor recurrence, providing important metabolic information [72]. In patients with suspected ORNJ, PET/CT findings often predominate in soft tissues, where recurrences are more likely to occur because of the richer vascular network compared with bone. Nevertheless, malignant cells adjacent to bone may stimulate osteoclastogenesis, thereby promoting bone resorption and tumor spread. Despite these advantages, PET/CT has important limitations and is not widely used as a primary imaging modality for ORNJ assessment [95].

Early ORNJ may not be detectable on conventional radiographs. In advanced stages, panoramic radiography often shows poorly defined radiolucencies [91]. Imaging findings may range from apparently normal bone to pathological fractures, localized or extensive osteolytic areas, and bone sequestra [72]. In the early stages, radiopaque or mixed radiolucent-radiopaque changes may also be observed, reflecting bone degradation and inflammatory alterations [72]. Compared with panoramic radiography, CT provides superior spatial resolution and enables more accurate detection of cortical disruption, trabecular loss, lytic areas, and soft tissue thickening. For this reason, CT is recommended for both the diagnosis and monitoring of ORNJ [90]. In addition, CT has shown an accuracy rate of 90% in differentiating ORNJ from other diseases and allows more precise characterization of necrotic areas [72].

CBCT is generally recommended when panoramic radiography provides insufficient diagnostic information, as it enables volumetric assessment of craniofacial bone structures at lower radiation exposure and cost than conventional CT [72]. It also provides detailed visualization of lesion morphology and extent, supporting the differential diagnosis of osteomyelitis, cystic lesions, tumors, and ORNJ. Conversely, CT offers superior soft tissue contrast, which facilitates the identification of infectious processes, particularly through the detection of gas bubbles. The presence of abnormal soft tissue proliferation in the affected region should raise suspicion for a second primary tumor or tumor recurrence [72].

The primary methodological limitations identified included inadequate allocation procedures, unclear inclusion criteria, absence of sample size calculation, and insufficient reporting of statistical assumptions. Such limitations compromise internal validity and may inflate effect estimates [96–101].

Substantial statistical heterogeneity was observed. Although sensitivity analyses were conducted, heterogeneity persisted, likely reflecting variability in radiation protocols, patient characteristics, outcome definitions, and study design rather than solely the number of included studies [26].

A major limitation of the present systematic review is that most included studies were retrospective cohort studies. Consequently, the available evidence should be interpreted considering the inherent limitations of retrospective data collection, including potential selection bias, information bias, incomplete clinical records, loss to follow-up, and residual confounding. The reliance on previously recorded clinical and histopathological data may also have led to misclassification of exposures, outcomes, or covariates. Moreover, differences in follow-up duration and clinical surveillance across studies may have influenced the detection of disease progression or malignant transformation [102, 103]. Substantial heterogeneity was also observed in some meta-analyses, likely reflecting variability in radiotherapy techniques, patient populations, ORNJ definitions, study designs, and imaging modalities used for ORNJ diagnosis as well as, interobserver variability. Importantly, the excessive heterogeneity observed in some meta-analyses was one of the main factors contributing to the downgrading of the certainty of evidence in the GRADE assessment.